Comparison of downscaling methods for mean and extreme precipitation in Senegal

HAL Id: hal-01871883

https://hal.umontpellier.fr/hal-01871883

Submitted on 7 Jun 2021

HAL is a multi-disciplinary open access

archive for the deposit and dissemination of

sci-entific research documents, whether they are

pub-lished or not. The documents may come from

teaching and research institutions in France or

abroad, or from public or private research centers.

L’archive ouverte pluridisciplinaire HAL, est

destinée au dépôt et à la diffusion de documents

scientifiques de niveau recherche, publiés ou non,

émanant des établissements d’enseignement et de

recherche français ou étrangers, des laboratoires

publics ou privés.

Distributed under a Creative Commons Attribution| 4.0 International License

Comparison of downscaling methods for mean and

extreme precipitation in Senegal

M.A. Sarr, O. Seidou, Y. Tramblay, S. El Adlouni

To cite this version:

M.A. Sarr, O. Seidou, Y. Tramblay, S. El Adlouni. Comparison of downscaling methods for mean and

extreme precipitation in Senegal. Journal of Hydrology: Regional Studies, Elsevier, 2015, 4, pp.369

-385. �10.1016/j.ejrh.2015.06.005�. �hal-01871883�

ContentslistsavailableatScienceDirect

Journal

of

Hydrology:

Regional

Studies

j ou rn a l h o m e pa g e : w w w . e l s e v i e r . c o m / l o c a t e / e j r h

Review

Comparison

of

downscaling

methods

for

mean

and

extreme

precipitation

in

Senegal

M.A.

Sarr

_,

_O.

_Seidou

_,

_Y.

_Tramblay

_,

_S.

_El

_Adlouni

a_{DépartementdeGéographie,UniversitédeMontréal,Canada} b_{DepartmentofCivilEngineering,UniversityofOttawa,Canada} c_{IRD,HydrosciencesMontpellier,France}

d_{DépartementdeMathématiquesetdeStatistique,UniversitédeMoncton,Canada}

a

r

t

i

c

l

e

i

n

f

o

Articlehistory:

Received2December2014

Receivedinrevisedform12May2015 Accepted8June2015

Availableonline4August2015 Keywords: Climatechange Extremeprecipitation Downscaling Quantile–quantiletransformation Delta-change

a

b

s

t

r

a

c

t

Studyregion:Thestudyconsiderssixprecipitationstationslocatedin Sene-gal,WestAfrica.SenegalislocatedintheSahel,anareathatisthreatenedby climatevariabilityandchange.Bothdroughtsandextremerainfallhavebeen anissueinrecentyears.

Studyfocus:Twodifferentstatisticaldownscalingtechniqueswereapplied totheoutputsoffourregionalclimatemodelsatsixselectedprecipitation stationsinSenegal.First,thedelta-changemethodwasappliedtothemean annualprecipitationaswellasthe5,10,20,50and100-yearreturnperiod dailyprecipitationevents.Second,aquantile–quantiletransformation(QQ) wasusedtodownscalethemonthlydistributionsofprecipitationsimulated byregionalclimatemodels(RCMs).The5,10,20,50and100-yeardaily precip-itationeventswereafterwardcalculated.Allextremeeventswerecalculated assumingthatmaximumannualdailyprecipitationsfollowthegeneralized extremevalue(GEV)distribution.Thetwo-sidedKolmogorov–Smirnov(KS) testwasfinallyusedtoassesstheperformanceofthequantile–quantile trans-formationaswellastheGEVdistributionfitfortheannualmaximumdaily precipitation.

Newhydrologicalinsightsfortheregion:Resultsshowthatthetwo down-scalingtechniquesgenerallyagreeonthedirectionofthechangewhenapplied totheoutputsofsameRCM,butsomecasesleadtoverydifferentprojections ofthedirectionandmagnitudeofthechange.Projectedchangesindicatea declineinmeanprecipitationexceptforoneRCMoveroneregioninSenegal. Projectedchangesinextremeprecipitationsarenotconsistentacrossstations andreturnperiods.Thechoiceofthedownscalingtechniquehasmoreeffect ontheestimationofextremedailyprecipitationsofreturnperiodequalor greaterthantenyearsthanthechoiceoftheclimatemodels.

©2015TheAuthors.PublishedbyElsevierB.V.Thisisanopenaccessarticle undertheCCBYlicense(http://creativecommons.org/licenses/by/4.0/).

∗ Correspondingauthor.Tel.:+15147080072.

E-mailaddresses:mamadousarr74@hotmail.com(M.A.Sarr),ousmane.seidou@uottawa.ca(O.Seidou), yves.tramblay@ird.fr(Y.Tramblay),Salah-eddine.el.adlouni@umoncton.ca(S.ElAdlouni).

http://dx.doi.org/10.1016/j.ejrh.2015.06.005

2214-5818/© 2015 The Authors.Published by ElsevierB.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).

Contents

1. Introduction...370

2. Methodology...372

2.1. Observedprecipitation ... 372

2.2. Regional/globalclimatemodelscombinations...372

2.3. T-yearprecipitationestimate...373

2.4. Thedelta-changetechnique...374

2.5. Thequantile–quantiletransformation(QQ)...374

3. Resultsanddiscussion...375

3.1. Validationofthequantile–quantiletransformationinthehistoricalperiod...375

3.2. Downscalingresults ... 376 3.2.1. Dakar...382 3.2.2. Linguère...382 3.2.3. Podor...382 3.2.4. Ziguinchor...382 3.3. Discussion...382 4. Conclusion...384 Conflictofinterest...384 References...384 1. Introduction

Senegalisasemi-aridcountrylocatedintheextremewestoftheAfricancontinent.Itsclimate ischaracterizedbyadryseasonfromNovembertoAprilandawetseasonfromMaytoOctober, regulatedbythemigrationsoftheInterTropicalConvergenceZone(ITCZ)(Leroux,1970).Duringthe wetseason,theITCZcompletelycoversthestudyareaasitmovestowardsthenorthtotheedge oftheSaharauntilthemonthsofAugustandSeptember.MostoftherainfallisrecordedinAugust, followedbyJulyandthenSeptember(Sarretal.,2013,2014).AspartoftheSahelianzone,Senegal hasbeenhitbyrecurrentdroughtsleadingtofoodshortage.Recently,thecountryalsoexperienced frequentepisodesofintenseprecipitationsleadingtourbanfloodingand casualtiesinDakar,the country’scapital,andinotherurbancentres.BetweenAugust16andAugust22,2005,Dakarhas recorded367mmofrain,whichismorethanhalfoftheaverageannualtotalrainfall.OnAugust26, 2012,Dakarreceivedonequarteroftheannualprecipitation(168mm)inlessthananhour(Dacosta, 2012,personalcommunication).Theseprecipitationeventshavecausedconsiderablepsycho-social andhealthimpacts,whichareeventuallossestovulnerablesectorssuchasagriculture,infrastructures andtrades.Theperceivedincreaseinclimate-relateddisastershastriggeredfearsthatthefrequency ofextremeprecipitationswillbetriggeringmoreextremeevents,characterizedbytheirinfrequency andtheirhighmagnitude.

Thebestwaytodevelopcost-effectivecopingstrategiesistogenerateinformationaboutfuture changesinmeanandextremeprecipitationindices(NichollsandMurray,1999;Mantonetal.,2001). Theseindicesaregenerallycalculatedwiththeoutputsofaclimatemodelrunningunderagiven scenario.Different emission scenarios and climatemodelssimulationsare available.In addition, differentdownscalingtechniquesexist,rangingfromstatisticalmethods(delta-change, regression-based,weathertyping,neuralnetworks,etc.)totheuseofregionalclimatemodels(RCMs).Statistical downscalingcanbeappliedtoRCMsinanattempttocorrecttheirbiases.Giventhatforpractical reasonsanenduserina givenregioncanonlyapplyafewoftheavailabletechniquesavailable, inter-comparison studies arehelpful in pointing out theuncertainty associated with thechoice ofthedownscalingtechnique.Burgeretal.(2012)showedforthesamelocationthatdownscaled climateextremesaremoresensitivetothechoiceofthedownscalingtechniquethantheemission scenario,theRCMandthelocation.Thechoiceofthedownscalingtechnique,mainlyinthefield ofregionalplanninganddecision-making,isveryimportant.Eveniftherearemanydownscaling techniques, theycan be classifiedinto two categories:dynamical and statistical (Hewitson and Crane,1996;Miller etal.,2009).Dynamicaldownscaling isphysicallyrelatedtoregionalclimate

Fig.1.Thestudydomainandtheselectedstations.

modelswithhighcomputationalrequirements(e.g.Giorgietal.,1994;Syllaetal.,2009).Dynamical downscalingdonotgivesignificantlybetterresultsfortemperatureandprecipitationandisoften consideredtooexpensiveforoperationaluse,butitdoesnotrequirelocalcalibrationoftheiroutputs. Statisticaldownscalingassumesstationarityinthepredictor–predictandrelationship,requiresrobust relationshipsandsufficientdatatoverifythisassumption.Statisticaldownscalingresultscouldbe alsostrongly impactedbydata errorsif thepredictor–predictandrelationship doesnotconsider importantclimaticfeatures(suchaslargescalecirculationandlocalcharacteristicsofthestudyarea), itgenerallyfocusesonprecipitation(inthepresentpaper)and/ortemperature(e.g.DiVittorioand Miller,2013).StatisticalmethodsarebasedonstatisticalrelationshipsbetweenthecoarseGCMor RCMdataandpointmeasurementatanobservationstation.Inpracticalterms,thestatistical rela-tionshipsarebasedonacalibrationperiod,approvedforaperiodofseparatetime,andthenapplied toothertimeperiods,withtheassumptionoftemporalstationarity(DiVittorioandMiller,2013).

The objective of this paper is to compare the impact of two downscaling techniques (the delta-changeandquantile–quantiletransformation)onprojectedchanges,inaverageandextreme precipitationofthe2000–2050periodatsixlocationsacrossSenegal(Fig.1).FourRCMsandglobal climatemodels(GCMs)aresampledfromEnsembles-basedPredictionsofClimateChangesandTheir Impacts(ENSEMBLES)andAfricanMonsoonMultidisciplinaryAnalyses(AMMA)experiments(Van derLindenandMitchell,2009;Redelspergeretal.,2006).Thedownscalingtechniqueswereapplied toestimatethemeandailyprecipitationand5,10,20,50and100-yeardailyprecipitationevents atthesixlocationsinadditiontothedirectionofchange.Thedelta-changetechniquewasselected becauseitismostwidelyusedwithRCMoutputs(Maraunetal.,2010;Themeßletal.,2011).Thedelta changetechniqueiseasytoapply(onejustneedtoapplyacoefficienttohistoricaltimeseries)and preserveimportantstatisticalcharacteristics(spatialcorrelations,interdependence)indownscaled timeseries.Itsmainlimitationisthatitdoesnotallowforachangeinvarianceinthefuture,and missestheriskgeneratedbyapossiblechangeinclimatevariablesdistributionsinthefuture.Itis

Table1

Selectedclimatestations.

ID Name Latitude Longitude Altitude(m) Dataavailability

BK Bakel 14.9 −12.5 25 1950–2007 DK Dakar 14.7 −17.5 24 – LG Linguère 15.4 −15.1 20 – KG Kédougou 12.6 −12.2 116 – PD Podor 16.6 −14.9 6 – ZG Ziguinchor 12.5 −16.3 26 –

wellknownthatthecorrectreproductionofthesestatisticalcharacteristicsisrequiredforarealistic

modellingofbiophysicalimpacts.ThechoiceoftheQQcorrectionismotivatedbythefactthatittries

tocorrectthedistributionsofthevariables,thereforepotentiallyreducingthebiasofextremeevents

estimatorsinclimatemodelsimulations.WhenitisappliedtoseveralvariablesgeneratedbyanRCM,

thespatialcorrelationandinterdependenceinRCMoutputsareinheritedbydownscaledtimeseries.

RegressionbaseddownscalingtechniquessuchastheStatisticalDownScalingModel(SDSM)were

notusedgiventheirreportedlowexplainedvarianceinreproducingdailyprecipitationstatistics

dur-ingthehistoricalperiod(reportedlyfrom6%to45%):Wilbyetal.(1998),WilbyandDawson(2008)

andNguyenetal.(2004).Thedelta-changewasappliedtoboththemeanandextremeprecipitations indicescalculatedusinghistoricaldata;TheQQ-transformationwasappliedtohistoricaldatabefore themeanandextremeprecipitationindiceswerecalculatedforfutureperiods.

Theremainderofthepaperisorganizedasfollows:thetwodownscalingmethodsaredescribed inSection2.TheresultsarediscussedinSection3,andaconclusionisfinallygiveninSection4.

2. Methodology

2.1. Observedprecipitation

Dailyprecipitationtimeseriesspanningfrom1950to2007atsixclimaticstationswerecollected fromtheSenegalesemeteorologicalserviceforthestudy(Table1).Thestationswereselectedto provideagoodspatialcoverageofthedifferentclimaticzonesacrossSenegal:Dakar(DK)inthewest andZiguinchor(ZG)inthesouthwestarelocatedonthefringeoftheAtlanticOcean.Podor(PD)and Linguère(LG)inthenorthofthecountry;Bakel(BK)intheeastisrepresentativeofthemorearidareas, whileKédougou(KG)atthesoutheastexperiencesarelativelywetclimate(Fig.2).StationsinSenegal holdtheoldestdatainWestAfrica.However,thesedataaremissing,inconsistentorerroneousbecause ofabadmanagementofmaterialsandstaffwhomadethemeasurements.Onlystationscontainingless than2%ofmissingvalueswereretainedinthestudy.Manyotherstationsarenotconsideredbecause manyyearsweremissing.TherainyseasoninSenegalgenerallyextendsfromMaytoOctober,with theheaviestprecipitationsoccurringbetweenJulyandSeptember.Adecreasingtrendwasdetected inthemeanannualprecipitationatallstationsusingaMann–Kendalltestat95%ofconfidencelevel, butnotrendwasfoundinmaximumdailyprecipitations.

2.2. Regional/globalclimatemodelscombinations

AMMA-ENSEMBLESisaninternationalcollaborativemodelintercomparisonexperimentthat pro-videsasetofRCMsimulationsresultscoveringmostoftheAfricancontinentataresolutionof50km (PaethandDiederich,2011a,b).TheRCMsaredrivenbyeithertheERA-INTERIMreanalysis,orby globalclimatemodelsoutputssimulatedundertheSRESA1Bemissionscenarios(PaethandDiederich, 2011a,b).Thedatasetsareavailableforfreeintheonlinedatabase(http://ensemblesrt3.dmi.dk/).Four differentRCM/GCMcombinations(listedinTable2)areselectedfromtheAMMA-ENSEMBLESdatabase toincludethreedifferentRCMs(HIRHAM,REMOandRCA)drivenbytwodifferentGCMs(HADCM3 andECHAM5)(PaethandDiederich,2011a,b).AllRCMsarerunundertheSRESA1Bscenarios.The simulatedprecipitationtimeseriesofthefourGCM/RCMscombinationsareextractedforeachofthe sixcitiesandusedintheanalysis.Theuseofanensembleofdifferentclimatesimulationsallows

Fig.2.Time-seriesoftheannualprecipitationextremesforeachstation.

theevaluationoftheuncertaintiesintheindividualprojectionscomingfromtheRCM/GCMmodels combination.

2.3. T-yearprecipitationestimate

Inthisstudy,thegeneralizedextremevaluedistribution(GEV)isusedtomodelmaximumdaily precipitations.TheGEVdistributionhasaninterestingasymptoticalbehaviourandiscommonlyused tomodelextremeevents.Theextremevaluetheory(EVT)providestheconvergenceoftheextremesto theGEVdistributionsdespitetheoriginaldistributionofthedailydata.TheGEVdistributioncombines theGumbel,FréchetandWeibulldistributionsofextremevalues(Jenkinson,1955):

F(x)=

⎧

⎪

⎪

⎨

⎪

⎪

⎩













where ,˛and arethelocation,thescaleandtheshapeparameters,respectively.When<0, theGEVcorrespondstotheFréchetdistributionandcanbefittedtoheavytailedbehaviour.The>0 representstheWeibuldistributionandisusedtofitleftskewedsamples.Thecase,=0correspondsto theGumbeldistributionandhasmoderaterighttailwithtwoparameters,scaleandlocation(Katzetal.,

Table2

SelectedGCM/RCMcombinations.

Acronym RCM DrivingGCM Simulationperiod

DMI-HIRHAM5 HIRHAM5 ECHAM5-r3 1989–2050

MPI-M-REMO REMO ECHAM5-r3 1950–2050

METNOHIRHAM RCA HadCM3-Q0 1990–2050

2002).TheGEVparametersareestimatedwiththegeneralizedmaximumlikelihood(GML)method. TheGMLmethodisbasedonthesameprincipleasthemaximumlikelihood(ML)methodwithan additionalconstraintontheshapeparameter,toeliminatepotentiallyinvalidvalues(ElAdlounietal., 2007).Apriordistributionofadaptedtohydro-meteorologicalserieswasintroducedbyMartins andStedinger(2000).ThepriorforhasaBetadistribution(withshapeparameters˛=3andˇ=6) definedontheinterval[−0.5,+0.5],withamodeat−0.12.

2.4. Thedelta-changetechnique

Thetechniqueisbasedontheratiobetweentheperiods2001–2050and1950–2000oftheT-year precipitationcalculatedusingRCMsimulations.ForagivenstationandagivenRCM,thedelta-change techniqueisappliedinasimilarwayasinTramblayetal.(2012)usingthefollowingsteps:

1.Maximumdailyprecipitationsofeachyearinthehistoricalprecipitationtimeseriesareextracted fromtheobserveddata.

2.TheGEVdistributionisfittedtothesamplestocalculatetheT-yearreturnperiodprecipitation (PT,hist,obs)forT=5,10,20,50,100yearsusingthemaximumlikelihoodestimator.

3.MaximumdailyprecipitationsofeachyearintheRCM-simulatedprecipitationtimeseriesforthe historicalperiodareextracted.

4.ThemeandailyprecipitationandtheT-yearreturnperiodprecipitation(PT,hist,RCM)forT=5,10,20, 50,100yearsarecalculatedusingtheGEVdistributionforwhichparametersareobtainedusing theirmaximumlikelihoodestimator.

5.MaximumdailyprecipitationsofeachyearintheRCM-simulatedprecipitationtimeseriesforthe futureperiodareextracted.

6.ThemeandailyprecipitationandtheT-yearreturnperiodprecipitation(PT,hist,RCM)forT=5,10,20, 50,100yearsarecalculatedusingtheGEVdistributionwhich’sparametersareobtainedusingtheir maximumlikelihoodestimator.

7.ThefutureT-yearprecipitationisestimatedas PT,fut,delta=PT,fut,RCMPT,hist,obs

PT,hist,RCM (2)

2.5. Thequantile–quantiletransformation(QQ)

Thequantile–quantiletransformation(alsocalledquantilemapping)(Maraunetal.,2010;Themeßl etal.,2011)wasappliedonamonthlyintegralstoobtainstatisticaldistributionsofagivenclimate variableascloseaspossibletothestatisticaldistributionoftheobservedvariableonthehistorical period.Theprocedureforagivenmonthandprecipitationvariableisexplainedbelow:

1.Thehistoricaldataissplitinacalibrationperiod(1950–1988)andavalidationperiod(1989–2007). ThedailytimeseriesofthemonthareextractedforbothperiodsfrombothobservationsandRCM simulations

2.AnempiricalcumulativedistributionfunctionFobsisdevelopedusingtheobservationsonthe cali-brationperiod;anothercumulativedistributionfunctionFRCMisdevelopedusingtheRCMoutputs onthecalibrationperiod.

3.CorrectedRCMsimulationsaregeneratedonthevalidationperiodandfutureperiodsusingthe followingtransformation:XCORR=F_obs−1(FRCM(XRCM))whereXRCMisthevariableextractedfromraw RCMsimulationsandXCORRisthecorrectedvariable.

4.Theprobabilitymassfunction(PMF)ofprecipitationoccurrence(definedasintensity>1mm/day) aswellasthePDFofprecipitationintensityonrainydaysarebuilt.IfthePDF(orPMF)ofcorrected variableisclosertotheprobabilitydensityfunction(PDF)oftheobservationsthanthePDF(or PMF)ofthenon-correctedvariable,thequantile–quantiletransformationisappliedtofutureRCM simulationsofthatparticularvariable.

Table3

Combinationofstationsandmodelsselectedtocarry-onextremeprecipitationestimationafterQQ-transformation(OK=the stationisselected).

Model Stations

Bakel Dakar Linguère Kédougou Podor Ziguinchor

DMI-HIRHAM5 OK OK

MPI-M-REMO OK OK

METNOHIRHAM OK OK

INMRCA3 OK

5.Atwo-sampleKolmogorov–Smirnov(KS)testisusedtocomparethePDFoftheobservationsto

thePDFofthecorrectedprecipitationonboththecalibrationandvalidationperiod(Simardand

L’Ecuyer,2011).Thenullhypothesisisthatthetwodatasetsarefromthesamecontinuous dis-tribution.Thealternativehypothesisisthattheyarefromdifferentcontinuousdistributions.The distributionsareassumeddifferentifthep-valueisbelow5%.

3. Resultsanddiscussion

3.1. Validationofthequantile–quantiletransformationinthehistoricalperiod

ThehypothesisthattheuncorrectedRCMoutputshavethesamedistributionastheobservations wassystematicallyrejectedbytheKStestatallstationsonboththecalibrationandvalidationperiod, supportingtheassumptionthatadistributioncorrectionisneeded.Asexpectedthehypothesisthatthe correctedRCMoutputshavethesamedistributionasobservedwasneverrejectedonthecalibration periodacrossallvariables andstations.TheQQ-downscalingiscarriedonatastationonlyifthe nullhypothesisisnotrejectedforanyofthethreemonthsoftherainyseasonwherethemaximum precipitationoccursinSenegal(July,AugustandSeptember).Resultsshowthatthenullhypothesis washoweverrejectedonthevalidationperiodatsomeofthestationsforsomeofthemonths,andthat therateofrejectionvariedgreatlyacrossmodelsandstations:forinstance,nomodelwasretained forthecitiesofBakelandKédougoubecauseforeachofthemodelsunderconsideration,thenull hypothesiswasrejectedforatleastonemonthintherainyseason(Table3);onlymodelDMI-HIRHAM5 wasretainedforDakar,andonlyonemodel(outoffour)wasrejectedforthestationofLinguère.The completelistofmodel-stationcombinationsretainedtocarry-ontheQQ-transformationcanbefound inTable3.NoRCMisselectedatthetwoeasternstationsofBakelandZiguinchor,whileonlyoneis selectedatDakar,twoatPodorandZiguinchorandthreeatPodor(Table3).NoRCMisselectedat morethantwostations,makingitdifficulttodrawageneralconclusionaboutpotentialzonesofbetter performancesfortheclimatemodels.ThereforewecannotrecommendaparticularRCMforSenegal basedonthisstudy.Therejectionsmaybeduetoundetectedanomaliessuchasoutliersineitherthe observeddata,ortheRCMdata.

TheQQ-transformationalwaysimprovedthefitbetweenRCMsimulationsandobservation.Even whentheKStestledtotherejectionofthenull hypothesisonthevalidationperiod,thep-value obtained when comparingcorrected RCMoutputs toobservations washigher than the p-value obtainedwhencomparinguncorrectedRCMvaluestoobservations.Anexampleofsuch improve-mentispresentedinFig.3whereprobabilityofrainfalloccurrenceatDakarforeachmonthfrom JunetoOctoberplottedfortheobservations,aRCM(METNOHIRHAMinthiscase)outputsand QQ-transformedRCMoutputs.Thegraphsarepresentedforthecalibration(leftpanels)andvalidation (rightpanels)periods.Thedarkbluebarsrepresentstheprecipitationoccurrenceprobabilityforthe monthintheobservations;thelightbluebarsrepresenttheprecipitationoccurrenceprobability calcu-latedusingtheuncorrectedoutputsoftheRCM;thegreenbarsrepresenttheprecipitationoccurrence probabilitycalculatedusingtheQQ-correctedoutputsoftheRCM;theorangeandredbarspresent thesameresultsfortheuncorrectedandcorrectedRCMonthe2000–2050period.Itcaneasilybeseen thatthegreenbars(correctedRCM)arealwaysclosertothedarkbluebars(observations)thanthe lightbluebars(uncorrectedRCM).Furthermore,theprojectedrainfalloccurrenceprobabilitiesforthe

Fig.3. ProbabilitymassfunctionofprecipitationoccurrenceforDakar,RCMMETNOHIRHAM.

2000–2050periodaresignificantlydifferentofthoseoftheuncorrectedRCM.Theprobabilitydensity ofwetdaysprecipitationsforthesamecity(Dakar)andthesameRCM(Fig.4).Onthatfigure,the thickbluelinesrepresenttheobservations;thedottedgreenlinesrepresenttheuncorrectedRCM; thedashedorangelinesrepresenttheQQ-correctedRCM.Thecontinuousgreenlineandthedash-dot violetlinesrepresenttheuncorrectedandcorrectedRCMonthe2000–2050period.Onceagain,itcan beseenthatthedistributionofthecorrectedRCMoutputsarecloserthatoftheobservationsthanthe distributionoftheuncorrectedRCMoutputs.

3.2. Downscalingresults

Byconstruction,noperformancecanbecalculatedforthedelta-change,sinceitisthemodification oftheobserveddataaccordingtoaclimatechangesignal;theQQtechniqueintheotherhandallows themodellertoevaluateitsabilitytocorrectprobabilitydistributions.Resultsshowthatthetwosided KStestleadtothediscardingasignificantlyhighnumberofstations-modelcombinations(Table3). InsimilarapplicationsoftheQQtransformation(e.g.Amadouetal.,2014;El-Khouryetal.,2015),the authorshavefoundthatthetransformationworksmuchbetterwithtemperaturesthanwith precipita-tion.Itisassumedthatthedifferenceinperformanceisduetothehighlyskewedshapeofprecipitation transformation,theQQtransformationhavingtroublereproducingthetailofthedistribution,sinceit isanon-parametricmethodthusheavilydependentonthecalibrationperiod.Themeanprecipitation andtheextremeprecipitationsofthehistorical(1950–2000)andfuture(2001–2050)periodsforthe

Fig.4.EmpiricalprobabilitydensityofwetdaysprecipitationforDakar,RCMMETNOHIRHAM.

modelsthatpassedtheKS-testscreeningarepresentedinTable4(Dakar),Table5(Linguère),Table6 (Podor)andTable7(Zinguichor).Eachtableisorganizedasfollow:thefirstandsecondcolumns rep-resenttheclimatemodelandtheindices(eitherthemeanprecipitationoraT-yearprecipitation). Columns3–8representthefollowingmethodsofestimationoftheindices:GEVfitonobservations (column3),GEVfitonuncorrectedclimatemodeloutputsonthehistoricalperiod(columns4),GEVfit onQQ-correctedclimatemodelsoutputsonthehistoricalperiod(columns5),GEVfitonuncorrected climatemodeloutputsonthefutureperiod(columns6),GEVfitonQQ-correctedclimatemodels out-putsonthefutureperiod(columns7)andfinallythedelta-changetechnique(column8).Eachclimate modelisrepresentedbysevenlines.TheresultsoftheKS-testcomparingthefittedGEVdistribution totheempiricaldistributionofannualmaximumprecipitationarepresentedonthefirstline.When ‘S.D.’(similardistribution)appearsinonecolumn,itmeansthattheKS-testdidnotleadtothe rejec-tionofthenullhypothesis.Thep-valueispresentedaswell.Thesecondlinepresentsthemeandaily precipitationwhilethethirdtotheseventhlinesrepresenttheT-yearquantileforT=5,10,20,50and 100years,andthecolumnrepresentsafutureperiod(columns7and80,apercentageofincreaseover thesamequantitycalculatedwiththeobservationsisprovided).Thefollowinggeneralconclusions canbedrawn:

1.TheannualmaximumdailyprecipitationtimeseriescanbeconsideredGEV-distributedsincethe nullhypothesiswasneverrejected.

2.Therangeofprojectedchangesrangesfromverysmallvalues(lessthan2%)withiswithinthe datauncertaintyrangetoquitelargevalues(from−41.6%to+18.7%).Whilesmallerchangesofa fewpercentscanbeconsiderednonsignificant,thoseabove10%stronglysuggestanimpactinthe future.

M.A. Sarr et al. / Journal of Hydrology: Regional Studies 4 (2015) 369–385 Table4

CurrentandfutureestimatesofextremedailyprecipitationsforDakar(M:annualprecipitationinmm;T=X:X-yearsreturnperiodprecipitation).

RCM/GCM Returnperiod Historicalperiod(1950–2000) Future(2001–2050)

Observations GCM/RCM GCM/RCM+QQ GCM/RCM(uncorrected) GCM/RCM+QQ GCM/RCM+delta-change KStest S.D.(p=0.879) S.D.(p=0.879) S.D.(p=0.411) S.D.(p=0.832) S.D.(p=0.926) METNOHIRHAM M 298.935 401.5 292 328.5 182.5(−39%) 255.5(−18.2%) T=5 86.5 103.1 73.2 88.0 55.9(−35.4%) 73.78(−14.7%) T=10 108.9 129.9 112.1 118.7 74.74(−31.3%) 99.41(−8.7%) T=20 130.1 157.8 174.5 154.9 98.28(−24.5%) 127.67(−1.9%) T=50 157.3 197.4 319.4 214.1 139.35(−11.4%) 170.64(+8.5%) T=100 177.4 229.8 509.9 269.9 180.52(+1.7%) 208.43(+17.5%)

M.A. Sarr et al. / Journal of Hydrology: Regional Studies 4 (2015) 369–385 379

CurrentandfutureestimatesofextremedailyprecipitationsforLinguère(M:annualprecipitationinmm;T=X:X-yearreturnperiodprecipitation). RCM/GCM Returnperiod Historicalperiod(1950–2000) Future(2001–2050)

Observations GCM/RCM GCM/RCM+QQ GCM/RCM(uncorrected) GCM/RCM+QQ GCM/RCM+delta-change

DMI-HIRHAM5 M 337.26 547.5 365 474.5 292(−13.5%) 292(−13.5%) T=5 76.8 42.4 80.2 39.8 74.17(−3.4%) 71.99(−6.2%) T=10 91.5 45.7 84.1 43.7 81.17(−11.2%) 87.41(−4.4%) T=20 108.2 48.4 86.3 47.1 86.01(−20.5%) 105.39(−2.6%) T=50 134.7 51.1 87.8 51.0 90.38(−32.9%) 134.33(−0.3%) T=100 158.8 52.8 88.4 53.6 92.65(−41.6%) 161.06(+1.5%) RCM/GCM KStest S.D.(p=0.879) S.D.(p=0.879) S.D.(p=0.640) S.D.(p=0.620) S.D.(p=0.617) MPI-M-REMO M 0.969 0.9 1.0 0.9 0.9(−7.2%) 0.6(−9.7%) T=5 83.5 123.5 80.1 112.9 75.88(−9.1%) 76.35(−8.5%) T=10 99.9 186.2 97.0 178.9 90.53(−9.4%) 96.03(−3.9%) T=20 115.2 269.6 113.8 274.5 104.95(−8.9%) 117.34(+1.8%) T=50 134.4 426.1 136.3 472.3 124.17(−7.6%) 148.94(+10.8%) T=100 148.3 594.3 153.8 705.5 139(−6.2%) 175.9(+18.7%) RCM/GCM KStest S.D.(p=0.928) S.D.(p=0.928) S.D.(p=0.902) S.D.(p=0.935) S.D.(p=0.959) METNOHIRHAM M 0.922 0.7 0.9 0.7 0.8(−13.3%) 0.9(−0.3%) T=5 74.6 70.2 87.0 72.3 82.15(+10.1%) 76.82(+3.0%) T=10 89.0 95.7 111.4 92.9 98.38(+10.6%) 86.3(−3.0%) T=20 105.7 127.6 139.3 115.3 113.88(+7.7%) 95.51(−9.6%) T=50 132.7 183.0 183.5 148.7 133.86(+0.9%) 107.81(−18.8%) T=100 157.8 238.4 223.9 177.4 148.78(−5.7%) 117.41(−25.6%)

M.A. Sarr et al. / Journal of Hydrology: Regional Studies 4 (2015) 369–385 Table6

CurrentandfutureestimatesofextremedailyprecipitationsforPodor(M:annualprecipitationinmm;T=X:X-yearsreturnperiodprecipitation). RCM/GCM Returnperiod Historicalperiod(1950–2000) Future(2001–2050)

Observations GCM/RCM GCM/RCM+QQ GCM/RCM(uncorrected) GCM/RCM+QQ GCM/RCM+delta-change KStest S.D.(p=0.449) S.D.(p=0.449) S.D.(p=0.690) S.D.(p=0.780) S.D.(p=0.297) DMI-HIRHAM5 M 210.605 255.5 182.5 182.5 109.5(−48.1%) 146(−29%) T=5 64.2 34.4 56.3 35.4 50.48(−21.3%) 65.84(+2.6%) T=10 69.6 35.9 63.5 39.6 58.51(−16.0%) 76.9(+10.5%) T=20 73.0 36.8 69.3 43.1 64.89(−11.1%) 85.52(+17.2%) T=50 75.6 37.4 75.3 46.7 71.58(−5.4%) 94.36(+24.7%) T=100 76.8 37.7 79.0 48.9 75.65(−1.6%) 99.61(+29.6%)

Sarr et al. / Journal of Hydrology: Regional Studies 4 (2015) 369–385 381 Table7

CurrentandfutureestimatesofextremedailyprecipitationsforZinguichor(M:annualprecipitationinmm;T=X:X-yearsreturnperiodprecipitation). RCM/GCM Returnperiod Historicalperiod(1950–2000) Future(2001–2050)

Observations GCM/RCM GCM/RCM+QQ GCM/RCM(uncorrected) GCM/RCM+QQ GCM/RCM+delta-change KStest S.D.(p=0.861) S.D.(p=0.86) S.D.(p=0.435) S.D.(p=0.842) S.D.(p=0.256) IMETNOHIRHAM M 1226.035 985.5 1131.5 985.5 1131.5(−7.8%) 1241(−0.3%) T=5 121.8 144.1 134.8 153.6 124.7(+2.4%) 129.87(+6.6%) T=10 148.0 186.7 138.9 189.8 134.4(−9.2%) 150.46(+1.7%) T=20 175.9 237.9 140.7 227.4 141.08(−19.8%) 168.11(−4.4%) T=50 216.7 323.5 141.5 280.4 147.08(−32.1%) 187.82(−13.3%) T=100 251.1 405.7 141.8 323.6 150.19(−40.2%) 200.3(−20.2%) RCM/GCM KStest S.D.(p=0.504) S.D.(p=0.50) S.D.(p=0.833) S.D.(p=0.960) S.D.(p=0.491) MPI-M-REMO M 3.581 2.8 3.5 3.0 3.6(0.5%) 3.8(6.9%) T=5 125.1 221.1 123.9 247.3 134.42(+7.5%) 139.91(+11.8%) T=10 150.9 280.2 138.4 291.2 143.21(−5.1%) 156.84(+4.0%) T=20 178.6 344.5 150.4 330.4 148.94(−16.6%) 171.3(−4.1%) T=50 219.4 440.4 163.8 377.1 153.77(−29.9%) 187.81(−14.4%) T=100 254.0 523.1 172.4 409.3 156.11(−38.5%) 198.74(−21.8%)

3.2.1. Dakar

OnlyoneRCM(METNOHIRHAM)wasselectedforthisstation.TheKS-testcomparingthefittedGEV

distributiontothatoftheannualmaximumdailyprecipitationdidnotleadtotherejectionofthe

null-hypothesisinanydataset.Thetwotechniquesresultsprojectareductionofthemagnitudeofboththe

meanprecipitationandextremeprecipitationeventswithareturnperiodequaltoorbelow20years,

buttheQQ-transformationprovidealargerdecreaseforthefuture(−39%versus−18.2%).According

totheQQ-transformation,the50-yeareventisalsosettodecreasewhilethedelta-changepredictsa

changeintheoppositedirection.Bothtechniquespredictanincreaseinthe100-yearevent,butthe

increasesuggestedbytheQQ-transformationistentimessmaller(+1.7%versus+17.5%).Giventhat

lowerquantilesaresettodecreaseandhigherquantilessettoincrease,thevariabilityofprecipitation

inDakarwillbeincreasing.

3.2.2. Linguère

ThreeRCMs(DMI-HIRHAM5,MPI-M-REMO,METNOHIRHAM)wereselectedforthisstation.The

twosidedKStestcomparingthefittedGEVdistributiontothatoftheannualmaximumdaily

precipi-tationdidnotleadtoasignificantdifferenceforalldataset.AllRCMsandbothdownscalingtechniques

provideareductionofthemeanprecipitationinarangebetween−13.5%and−0.3%.Thereishowever

adisagreementbetweenmodelsanddownscalingtechniquesaboutthedirectionsofchangesofthe

quantiles.WhenappliedtotheoutputsofDMI-HIRHAM5andMPI-M-REMO,theQQ-transformation

predictsadecreaseofthequantilesforallreturnperiodswhilethedelta-changetechniquepoints

toadecreaseinlowerreturnperiodsandanincreaseinhigherreturnperiods.Thetwodownscaling

techniquesagreeonthedirectionofchangeofthequantileswhentheyareappliedtotheoutputsof

METNOHIRHAM:theypointtoanincreaseinlowerreturnperiodsquantilesandadecreaseinhigher

returnperiodsquantiles.

3.2.3. Podor

TwoRCMs(INMRCA3andDMI-HIRHAM5)wereselectedforthisstation.ThetwosidedKS-test

comparingthefittedGEVdistributiontothatoftheannualmaximumdailyprecipitationdidnotlead

tothesameresultoftheabsenceofsignificantdifferences.AllRCMsandbothdownscalingtechniques

provideareductionofthemeanprecipitationinarangebetween−29%and−48.1%.Thedelta-change

suggestsanincreaseinallquantileswhiletheQQ-transformationsystematicallysuggestedachange

intheoppositedirection.

3.2.4. Ziguinchor

TwoRCMs(METNOHIRHAMandMPI-M-REMO)wereselectedforthisstation.METNOHIRHAM

predictedareductionofthemeanprecipitation(−7.8%withtheQQ-transformation;−0.3%withthe

delta-change)whileMPI-M-REMOpointstoanincreaseofthemeanprecipitation(+0.5%withthe

QQ-transformation;+6.9%withthedelta-change).Independentlyofthedownscalingtechniqueused,

bothclimatemodelsshowanincreaseoflowerreturnperiodsofthequantilesandadecreaseofhigher

returnperiodofthequantiles.

3.3. Discussion

Themostimportantfindingofthisstudyisthatforagivenlocation,dependingontheclimate

modelanddownscalingtechniqueused,onecangettwooppositeconclusions.Itisthereforehighly

hazardoustorelyononlyonemodeloronlyonedownscalingtechniquewhendesigningadaptation

optionstoclimatechangebasedonmodelsimulations.Asshownin previousstudies,combining

differentclimatemodelscanincreasetheconsistencyofmodelprojections(Freietal.,2006;Fowler

etal.,2007a;TebaldiandKnutti,2007;Déquéetal.,2012).However,itmustbenotedthatwhenthe sameRCMisusedwithbothtechniques,thedelta-changeandtheQQ-transformationalwaysagreed inthedirectionoftheprojectedchangeforthemeanannualprecipitation(predicteddownwardin sevenoutofeightcasesinthisstudy),butnotforthechangeinmagnitude.Ontheopposite,thetwo downscalingtechniquescandisagreeonthedirectionofchangeforextremeprecipitationquantiles, evenwhenappliedtothesameRCM.Severalauthorshavepreviouslyshownthatthedifferenceinthe

Table8

p-ValueandrankofeachfactoraftertheANOVAanalysis(boldvaluesrepresentssignificantinfluence;(M:annualprecipitation inmm;T=X:X-yearsreturnperiodsprecipitation).

Index City GCM RCM Downscalingtechnique

M 1.33e−007(rank=1) 5.99e−001(rank=3) 7.92e−001(rank=4) 4.37e−001(rank=2)

T=5 1.09e−005(rank=1) 2.47e−001(rank=4) 1.38e−001(rank=3) 8.27e−002(rank=2)

T=10 8.57e−005(rank=1) 1.31e−001(rank=4) 6.68e−002(rank=3) 1.93e−00s2(rank=2)

T=20 1.25e−003(rank=1) 1.01e−001(rank=4) 5.16e−002(rank=3) 9.83e−003(rank=2)

T=50 2.45e−002(rank=2) 1.02e−001(rank=4) 5.60e−002(rank=3) 8.50e−003(rank=1)

T=100 4.79e−002(rank=2) 1.24e−001(rank=4) 7.17e−002(rank=3) 1.08e−002(rank=1)

climatechangesignalbetweenbias-correctedanduncorrectedRCMscouldbelowformeanvalues

ofhydrologicalvariablessuchasprecipitation(Chenetal.,2011), butmaybehighinthecaseof

extremevalues(Hagemannetal.,2011;Dosioetal.,2012).Thisfindingisconfirmedinthepresent studywithRCM,indicatingthatfutureprojectionsofextremeeventsmaybeassociatedwithahigh uncertainty.ThemainsourcesofuncertaintyinregionalclimatechangearethechoiceoftheRCMand thechoiceoftheforcingGCM(Déquéetal.,2012).Inaddition,thebias-correctionofclimatemodel outputsreliesonthehypothesisofstationarityofthebias(Maraunetal.,2010).Severalauthorshave shownthatthishypothesiswasnotfulfilledforsomearidregions(Maraun,2012)orforconsidering differenttimeperiodsconcerningthevalidationandthecalibrationofthemethod(Lafonetal.,2013). GiventhatsomeclimatemodelshavebeendiscardedafterfailingtheKS-testinvalidation,theonly stationwheretheresultsfordifferentRCMs(HIRHAM5andREMO)drivenbythesameGCM(ECHAM5) canbecompared isLinguère(Table5).Thedifferencesbetweentheprojectedvariationsinmean precipitationislessaffectedbythechoiceofdownscalingtechniques(predictedchangesdifferedof 0.0%and2.5%ofthehistoricmeandependingoftheRCM)thanbythechoiceoftheRCM(6.2%and 3.8%ofthehistoric meandependingofthedownscalingtechnique).Asimilarcomparison canbe donebutmaybemisleadingforextremeprecipitationssincetheGEVfitonextremeprecipitationis calculatedwiththecorrectedoutputsofDMI-HIRHAM5atLinguèrewhichisoneoftheworstone inthedataset(withap-valueof0.574whilep-valuesforotherdatasetsareinthe0.9–1range). Finally,inordertoidentifywhich factorhasmoreimpactonthedownscaledmeanandextreme precipitationamongthelocation,thedrivingGCM,theRCMorthedownscalingtechnique,a four-waysfixedeffectANOVAanalysiswasperformed.Thelocationfactorhasfourlevels(Dakar,Podor, LinguèreandZiguinchor)correspondingtothecitiesforwhichonemodelwasretainedatleast.The GCMfactorhastwolevels(ECHAM5andHADCM3),theRCMfactorshave threelevels(HIRHAM, RCA,REMO)andfinallythedownscalingfactorhastwolevels(delta-changeandquantile–quantile transformation).Thep-valuesofthesignificanceofeach factorforeachoftheindices(meanand extremeprecipitations)andtheirrankinadecreasingorderofinfluencearelistedinTable8.Ata 95%confidencelevel,onlythelocationsignificantlyaffectsthemeanprecipitationandthe5-year returnperioddailyprecipitation.Thedownscalingbecomessignificantforreturnperiodsabove10 year.Ittakesthesecondrankafterthelocationfactorforreturnperiods10yearsand20years.The downscalingtechniquetakesthefirstrankandthelocationofthesecondrankforreturnperiods50 yearsand100years.TheGCMandtheRCMtechniquesdidnothaveasignificantimpact(p-value below0.05)onanyoftheindices,althoughthep-valueoftheRCMisalwayscloseto0.05which isthethresholdvalueforreturnperiodsabove10years.Theseresultsshouldbetakenwithcare asthesizeofthedatasetusedfortheANOVAissmall(16combinations).Resultssuggestthatone coulduseanyofthemodelsandanyofthedownscalingtechniquestoestimatethemeanand5-year precipitationatanyofthestations.ForT-yeardailyprecipitationwithTlargerorequalto10year, thechoiceofthedownscalingtechniquematters.Despitethefactthatrecentstudyshowedthatthe bias-correctionmethodcanimprovethereproductionofthehistoricalclimatebyRCMs,itishardto tellifonedownscalingtechniqueisbetterthantheother:severalstudies(TebaldiandKnutti,2007; ReifenandToumi,2009)mentionedthattheperformanceofclimatemodelsatreproducingthepast isnotaguaranteeofbetterskillsatmakingaccurateprojections.Ourrecommendationistouseboth inordertocapturetheuncertaintyassociatedwiththedownscalingtechnique.

4. Conclusion

ThisstudypresentsthefirstevaluationoftheprojectedchangesinprecipitationinSenegalwith RCMsandtwodownscalingtechniques.Thedailydataobservedatsixstationsbetween1950and 2007aswellasfourRCMsimulationsdrivenbytwodifferentGCMsaretakenintoaccounttoevaluate theimpactoffutureclimatechangesundertheemissionofgreenhousegasesscenarioA1B.GEV distributionsarefittedtotheprecipitationextremesobservedandprojectedtoprovideananalysis basedonquantilesfordifferentreturnperiods.Resultsshowthatthetwotechniquesgenerallyagree onthedirectionofthechangewhenappliedtotheoutputsofsameclimatemodel,butsometimeslead toverydifferentprojectionsofthedirectionandmagnitudeofthechangeinextremeprecipitations. ProjectedchangesinthemeanprecipitationaredownwardexceptforoneRCMinonecity.Projected changesinextremeprecipitationsarenotconsistentacrossstationsandreturnperiods.

Resultsalsosuggestthatthechoiceofthedownscalingtechniquehasmoreeffectontheestimation ofextremedailyprecipitationsofreturnperiodequalorgreaterthantenyearsthanthechoiceofthe climatemodels.Becauseofthegreatvariabilityinthefutureprojectionsobtainedwiththesetof4 RCMsconsideredherein,furtherworkinthesameregionshouldconsideralargerensembleofclimate modelstoevaluateifalargerensembleprovidessimilarresults.

Conflictofinterest

Nonedeclared.

References

Amadou,A.,Abdouramane,G.D.,Seidou,O.,SeidouSanda,I.,Sittichok,K.,2014.ChangestoflowregimeontheNigerRiverat Koulikorounderachangingclimate.Hydrol.Sci.J.,http://dx.doi.org/10.1080/02626667.2014.916407.

Burger,G.,Murdock,T.Q.,Werner,A.T.,Sobie,S.R.,Cannon,A.J.,2012.Downscalingextremes–anintercomparisonofmultiple statisticalmethodsforpresentclimate.J.Clim.25,4366–4388,http://dx.doi.org/10.1175/JCLI-D-11-00408.1.

Chen,C.,Haerter,J.O.,Hagemann,S.,Piani,C.,2011.Onthecontributionofstatisticalbiascorrectiontotheuncertaintyinthe projectedhydrologicalcycle.Geophys.Res.Lett.38,L20403,http://dx.doi.org/10.1029/2011GL049318.

Dacosta,A.,2012.Symposiuminternationalsur«Population,développementetchangementclimatique»,Dakar12–14décembre 2012(personalcommunication).

Déqué,M.,Somot,S.,Sanchez-Gomez,E.,Goodess,C.M.,Jacob,D.,Lenderink,G.,Christensen,O.B.,2012.Thespreadamongst ENSEMBLESregionalscenarios:regionalclimatemodels,drivinggeneralcirculationmodelsandinterannualvariability. Clim.Dyn.38,5–6.

DiVittorio,A.V.,Miller,N.L.,2013.Evaluatingamodifiedpoint-basedmethodtodownscalecell-basedclimatevariabledatato high-resolutiongrids.Theor.Appl.Climatol.112,495–519,http://dx.doi.org/10.1007/s00704-012-0740-9.

Dosio, A., Paruolo, P., Rojas, R., 2012. Bias correction of the ENSEMBLES high resolution climate change pro-jections for use by impact models: analysis of the climate change signal. J. Geophys. Res. 117, D17110, http://dx.doi.org/10.1029/2012JD017968.

ElAdlouni,S.,Ouarda,T.B.M.J.,Zhang,X.,Roy,R.,Bobée,B.,2007.Generalizedmaximumlikelihoodestimatorsforthe nonsta-tionarygeneralizedextremevaluemodel.WaterResour.Res.43,http://dx.doi.org/10.1029/2005WR004545.

El-Khoury,A.,Seidou,O.,Lapen,D.R.,Que,Z.,Mohammadi,M.,Sunohara,M.,2015.Combinedimpactsoffutureclimateand landusechangesondischarge,nitrogenandphosphorusloadsforaCanadianriverbasin.J.Environ.Manag.151,76–86, http://dx.doi.org/10.1016/j.jenvman.2014.12.012.

Frei,C.,Schöll,R.,Fukutome,S.,Schmidli,J.,Vidale,P.L.,2006.FuturechangeofprecipitationextremesinEurope: intercompar-isonofscenariosfromregionalclimatemodels.J.Geophys.Res.111,http://dx.doi.org/10.1029/2005JD005965.

Fowler,H.J.,Ekström,M.,Blenkinsop,S.,Smith,P.A.,2007a.EstimatingchangeinextremeEuropeanprecipitationusinga multimodelensemble.J.Geophys.Res.112,http://dx.doi.org/10.1029/2007JD008619.

Hagemann,S.,Cui,C.,Haerter,J.O.,Heinke,J.,Dieter,G.,Piani,C.,2011.Impactofastatisticalbiascorrectiononthe pro-jectedhydrologicalchangesobtainedfromthreeGCMsandtwohydrologymodels.J.Hydrometeorol.12,556–578, http://dx.doi.org/10.1175/2011JHM1336.1.

Hewitson,B.C.,Crane,R.G.,1996.Climatedownscaling:techniquesandapplication.Clim.Res.7(2),85–95.

Giorgi,F.,ShieldsBrodeur,C.,Bates,G.T.,1994.RegionalclimatechangescenariosovertheUnitedStatesproducedwithanested regionalclimatemodel.J.Clim.7(3),375–399.

Jenkinson,A.F.,1955.Thefrequencydistributionoftheannualmaximum(orminimum)ofmeteorologicalelements.Q.J.R. Meteorol.Soc.81,158–171.

Katz,R.W.,Parlange,M.B.,Naveau,P.,2002.Statisticsofextremesinhydrology.Adv.WaterResour.25,1287–1304. Lafon,T.,Dadson,S.,Buys,G.,Prudhomme,C.,2013.Biascorrectionofdailyprecipitationsimulatedbyaregionalclimatemodel:

acomparisonofmethods.Int.J.Climatol.33,1367–1381,http://dx.doi.org/10.1002/joc.3518.

Leroux,M.,1970.LadynamiquedesprecipitationsenAfriqueOccidentalePubl.Dir.Expl.Met.ASECNA,SerieI,no.23,Dakar, 121pp.

Manton,M.J.,etal.,2001.TrendinextremedailyrainfallandtemperatureinsoutheastAsiaandtheSouthPacific:1961–1998. Int.J.Climatol.21,269–284.

Maraun,D.,Wetterhall,F.,Ireson,A.M.,Chandler,R.E.,Kendon,E.J.,Widmann,M.,Brienen,S.,etal.,2010.Precipitation down-scalingunderclimatechange:recentdevelopmentstobridgethegapbetweendynamicalmodelsandtheenduser.Rev. Geophys.48,RG3003,http://dx.doi.org/10.1029/2009RG000314.

Maraun,D.,2012.NonstationaritiesofregionalclimatemodelbiasesinEuropeanseasonalmeantemperatureandprecipitation sums.Geophys.Res.Lett.39,L06706,http://dx.doi.org/10.1029/2012GL051210.

Martins,E.S.,Stedinger,J.R.,2000.GeneralizedmaximumlikelihoodGEVquantileestimatorsforhydrologicdata.WaterResour. Res.36,737–744.

Miller,N.L.,Cayan,D.,Duffy,P.,Hidalgo,H.,Jin,J.,Kanamaru,H.,Kanamitsu,M.,O’Brien,T.,Schlegel,N.,Sloan,L.,Snyder, M.,Yoshimura,K.,2009.AnAnalysisofSimulatedCAClimateusingMultipleDynamicalandStatisticalTechniques.The CaliforniaClimateChangeCenterReportSeries,CEC-500-2009-017-F.

Nicholls,N.,Murray,W.,1999.Workshoponindicesandindicatorsforclimateextremes:Asheville,NC,USA,3–6June1997 BreakoutGroupB:Precipitation.J.Clim.Change42(1),23–29,http://dx.doi.org/10.1023/A:1005495627778.

Nguyen,T.D.,Nguyen,V.T.V.,Gachon,P.,Bourque,A.,2004.Anassessmentofstatisticaldownscalingmethodsforgenerating dailyprecipitationandtemperatureextremesinthegreaterMontrealRegion.In:57thAnnualConferenceoftheCanadian WaterResourcesAssociation,Montreal,Québec,Canada,June16–18,2004,10pp.

Paeth,H.,Diederich,M.,2011a.PostprocessingofsimulatedprecipitationforimpactresearchinWestAfrica.PartI:model outputstatisticsformonthlydata.Clim.Dyn.36,1321–1336.

Paeth,H.,Diederich,M.,2011b.PostprocessingofsimulatedprecipitationforimpactresearchinWestAfrica.PartII:aweather generatorfordailydata.Clim.Dyn.36,1337–1348.

Redelsperger, J.-L., Thorncroft, C.D., Diedhiou, A., Lebel, T., Parker, D.J., Polcher, J.,2006. African monsoon multidis-ciplinary analysis: an international research project and field campaign. Bull. Am. Meteor.Soc. 87, 1739–1746, http://dx.doi.org/10.1175/BAMS-87-12-1739.

Reifen,C.,Toumi,R.,2009.Climateprojections:pastperformancenoguaranteeoffutureskill?Geophys.Res.Lett.36, http://dx.doi.org/10.1029/2009GL038082.

Sarr,M.A.,Gachon,P.,Seidou,O.,Bryant,C.R.,Ndione,J.,Comby,J.,2014.InconsistentlineartrendsinSenegaleserainfalldata indicesfrom1950to2007.Hydrol.Sci.J.,http://dx.doi.org/10.1080/02626667.2014.916407.

Sarr,M.A.,Zorome,M.,Seidou,O.,Bryant,C.R.,Gachon,P.,2013.Recenttrendsinselectedextremeprecipitationindicesin Senegal–achange-pointapproach.J.Hydrol.505,326–334.

Simard,R.,L’Ecuyer,P.,2011.Computingthetwo-sidedKolmogorov–Smirnovdistribution.J.Stat.Softw.VV(II).

Sylla,M.B.,Gaye,A.T.,Pal,J.S.,Jenkins,G.S.,Bi,X.Q.,2009.High-resolutionsimulationsofWestAfricanclimateusingregional climatemodel(RegCM3)withdifferentlateralboundaryconditions.Theor.Appl.Climatol.98,293–314.

Tebaldi,C.,Knutti,R.,2007.Theuseofthemulti-modelensembleinprobabilisticclimateprojections.Philos.Trans.R.Soc.Lond., Ser.A365,2053–2075.

Themeßl,M.,Gobiet,A.,Leuprecht,A.,2011.Empirical-statisticaldownscalinganderrorcorrectionofdailyprecipitationfrom regionalclimatemodels.Int.J.Climatol.31,1531–1544,http://dx.doi.org/10.1002/joc.2168.

Tramblay,Y.,Badi,W.,Driouech,F.,Neppel,El.S.,Servat,L.E.,2012.ClimatechangeimpactsonextremeprecipitationinMorocco. Glob.Planet.Change82–83,104–114.

VanderLinden,P.,Mitchell,J.F.B.(Eds.),2009.ENSEMBLES:ClimateChangeanditsImpacts.SummaryofResearchandResults fromtheENSEMBLESProject.MetOfficeHadleyCentre,Exete.

Wilby,R.L.,Hassan,H.,Hanaki,K.,1998.Statisticaldownscalingofhydrometeorologicalvariablesusinggeneralcirculation modeloutput.J.Hydrol.205,1–19.

Wilby,R.L.,Dawson,C.W.,2008.SDSM4.2–ADecisionSupportToolfortheAssessmentofRegionalClimateChangeImpacts. StatisticalDownScalingModelSDSMVersion4.2.